filaPro / oneformer3d

[CVPR2024] OneFormer3D: One Transformer for Unified Point Cloud Segmentation
Other
333 stars 33 forks source link

How should I make config file to train with my customized dataset? #70

Open HitmansGO opened 3 months ago

HitmansGO commented 3 months ago

I had already prepared my customized dataset with S3DIS format. And I also referred to mmdection3d's config file https://mmdetection3d.readthedocs.io/en/latest/user_guides/new_data_model.html but when I run train.py with my own config file, I got this erro:

Traceback (most recent call last):
  File "/home/hfut108/oneformer3d/tools/train.py", line 135, in <module>
    main()
  File "/home/hfut108/oneformer3d/tools/train.py", line 131, in main
    runner.train()
  File "/home/hfut108/anaconda3/envs/openmmlab3d/lib/python3.10/site-packages/mmengine/runner/runner.py", line 1777, in train
    model = self.train_loop.run()  # type: ignore
  File "/home/hfut108/anaconda3/envs/openmmlab3d/lib/python3.10/site-packages/mmengine/runner/loops.py", line 96, in run
    self.run_epoch()
  File "/home/hfut108/anaconda3/envs/openmmlab3d/lib/python3.10/site-packages/mmengine/runner/loops.py", line 112, in run_epoch
    for idx, data_batch in enumerate(self.dataloader):
  File "/home/hfut108/anaconda3/envs/openmmlab3d/lib/python3.10/site-packages/torch/utils/data/dataloader.py", line 681, in __next__
    data = self._next_data()
  File "/home/hfut108/anaconda3/envs/openmmlab3d/lib/python3.10/site-packages/torch/utils/data/dataloader.py", line 1376, in _next_data
    return self._process_data(data)
  File "/home/hfut108/anaconda3/envs/openmmlab3d/lib/python3.10/site-packages/torch/utils/data/dataloader.py", line 1402, in _process_data
    data.reraise()
  File "/home/hfut108/anaconda3/envs/openmmlab3d/lib/python3.10/site-packages/torch/_utils.py", line 461, in reraise
    raise exception
Exception: Caught Exception in DataLoader worker process 0.
Original Traceback (most recent call last):
  File "/home/hfut108/anaconda3/envs/openmmlab3d/lib/python3.10/site-packages/torch/utils/data/_utils/worker.py", line 302, in _worker_loop
    data = fetcher.fetch(index)
  File "/home/hfut108/anaconda3/envs/openmmlab3d/lib/python3.10/site-packages/torch/utils/data/_utils/fetch.py", line 49, in fetch
    data = [self.dataset[idx] for idx in possibly_batched_index]
  File "/home/hfut108/anaconda3/envs/openmmlab3d/lib/python3.10/site-packages/torch/utils/data/_utils/fetch.py", line 49, in <listcomp>
    data = [self.dataset[idx] for idx in possibly_batched_index]
  File "/home/hfut108/anaconda3/envs/openmmlab3d/lib/python3.10/site-packages/mmengine/dataset/dataset_wrapper.py", line 171, in __getitem__
    return self.datasets[dataset_idx][sample_idx]
  File "/home/hfut108/anaconda3/envs/openmmlab3d/lib/python3.10/site-packages/mmengine/dataset/base_dataset.py", line 418, in __getitem__
    raise Exception(f'Cannot find valid image after {self.max_refetch}! '
Exception: Cannot find valid image after 1000! Please check your image path and pipeline

and here is my config file:

import sys
sys.path.append("/home/hfut108")

_base_ = [
    'mmdet3d::_base_/default_runtime.py',
]
custom_imports = dict(imports=['oneformer3d.oneformer3d.oneformer3d'])

# model settings
num_channels = 64
num_instance_classes = 2
num_semantic_classes = 2
class_names = ['part','bgpart']
metainfo = dict(classes=class_names)
num_points = 4096

model = dict(
    type='S3DISOneFormer3D',
    data_preprocessor=dict(type='Det3DDataPreprocessor'),
    in_channels=6,
    num_channels=num_channels,
    voxel_size=0.05,
    num_classes=num_instance_classes,
    min_spatial_shape=128,
    backbone=dict(
        type='SpConvUNet',
        num_planes=[num_channels * (i + 1) for i in range(5)],
        return_blocks=True),
    decoder=dict(
        type='QueryDecoder',
        num_layers=3,
        num_classes=num_instance_classes,
        num_instance_queries=400,
        num_semantic_queries=num_semantic_classes,
        num_instance_classes=num_instance_classes,
        in_channels=num_channels,
        d_model=256,
        num_heads=8,
        hidden_dim=1024,
        dropout=0.0,
        activation_fn='gelu',
        iter_pred=True,
        attn_mask=True,
        fix_attention=True,
        objectness_flag=True),
    criterion=dict(
        type='S3DISUnifiedCriterion',
        num_semantic_classes=num_semantic_classes,
        sem_criterion=dict(
            type='S3DISSemanticCriterion',
            loss_weight=5.0),
        inst_criterion=dict(
            type='InstanceCriterion',
            matcher=dict(
                type='HungarianMatcher',
                costs=[
                    dict(type='QueryClassificationCost', weight=0.5),
                    dict(type='MaskBCECost', weight=1.0),
                    dict(type='MaskDiceCost', weight=1.0)]),
            loss_weight=[0.5, 1.0, 1.0, 0.5],
            num_classes=num_instance_classes,
            non_object_weight=0.05,
            fix_dice_loss_weight=True,
            iter_matcher=True,
            fix_mean_loss=True)),
    train_cfg=dict(),
    test_cfg=dict(
        topk_insts=450,
        inst_score_thr=0.0,
        pan_score_thr=0.4,
        npoint_thr=300,
        obj_normalization=True,
        obj_normalization_thr=0.01,
        sp_score_thr=0.15,
        nms=True,
        matrix_nms_kernel='linear',
        num_sem_cls=num_semantic_classes,
        stuff_cls=[1],
        thing_cls=[0]))

# dataset settings
dataset_type = 'S3DISSegDataset_'
data_root = 'data/s3dis/'
data_prefix = dict(
    pts='points',
    pts_instance_mask='instance_mask',
    pts_semantic_mask='semantic_mask')

train_area = [1, 2, 3, 4, 6]
test_area = 5

train_pipeline = [
    dict(
        type='LoadPointsFromFile',
        coord_type='DEPTH',
        shift_height=False,
        use_color=True,
        load_dim=6,
        use_dim=[0, 1, 2, 3, 4, 5]),
    dict(
        type='LoadAnnotations3D',
        with_label_3d=False,
        with_bbox_3d=False,
        with_mask_3d=True,
        with_seg_3d=True),
    dict(
        type='PointSample_',
        num_points=num_points),
    dict(type='PointInstClassMapping_',
        num_classes=num_instance_classes),
    dict(
        type='RandomFlip3D',
        sync_2d=False,
        flip_ratio_bev_horizontal=0.5,
        flip_ratio_bev_vertical=0.5),
    dict(
        type='GlobalRotScaleTrans',
        rot_range=[0.0, 0.0],
        scale_ratio_range=[0.9, 1.1],
        translation_std=[.1, .1, .1],
        shift_height=False),
    dict(
        type='NormalizePointsColor_',
        color_mean=[127.5, 127.5, 127.5]),
    dict(
        type='Pack3DDetInputs_',
        keys=[
            'points', 'gt_labels_3d',
            'pts_semantic_mask', 'pts_instance_mask'
        ])
]
test_pipeline = [
    dict(
        type='LoadPointsFromFile',
        coord_type='DEPTH',
        shift_height=False,
        use_color=True,
        load_dim=6,
        use_dim=[0, 1, 2, 3, 4, 5]),
    dict(
        type='LoadAnnotations3D',
        with_bbox_3d=False,
        with_label_3d=False,
        with_mask_3d=True,
        with_seg_3d=True),
    dict(
        type='MultiScaleFlipAug3D',
        img_scale=(1333, 800),
        pts_scale_ratio=1,
        flip=False,
        transforms=[
            dict(
                type='NormalizePointsColor_',
                color_mean=[127.5, 127.5, 127.5])]),
    dict(type='Pack3DDetInputs_', keys=['points'])
]

# run settings
train_dataloader = dict(
    batch_size=2,
    num_workers=3,
    persistent_workers=True,
    sampler=dict(type='DefaultSampler', shuffle=True),
    dataset=dict(
            type='ConcatDataset',
            datasets=([
                dict(
                    type=dataset_type,
                    data_root=data_root,
                    ann_file=f's3dis_infos_Area_{i}.pkl',
                    pipeline=train_pipeline,
                    metainfo=metainfo,
                    filter_empty_gt=True,
                    data_prefix=data_prefix,
                    box_type_3d='Depth',
                    backend_args=None) for i in train_area])))

val_dataloader = dict(
    batch_size=1,
    num_workers=1,
    persistent_workers=True,
    sampler=dict(type='DefaultSampler', shuffle=False),
    dataset=dict(
        type=dataset_type,
        data_root=data_root,
        ann_file=f's3dis_infos_Area_{test_area}.pkl',
        pipeline=test_pipeline,
        metainfo=metainfo,
        test_mode=True,
        data_prefix=data_prefix,
        box_type_3d='Depth',
        backend_args=None))
test_dataloader = val_dataloader

label2cat = {i: name for i, name in enumerate(class_names)}
metric_meta = dict(
    label2cat=label2cat,
    ignore_index=[num_semantic_classes],
    classes=class_names,
    dataset_name='S3DIS')
sem_mapping = [0, 1]

val_evaluator = dict(
    type='UnifiedSegMetric',
    stuff_class_inds=[1],
    thing_class_inds=[0],
    min_num_points=1,
    id_offset=2**16,
    sem_mapping=sem_mapping,
    inst_mapping=sem_mapping,
    submission_prefix_semantic=None,
    submission_prefix_instance=None,
    metric_meta=metric_meta)
test_evaluator = val_evaluator

optim_wrapper = dict(
    type='OptimWrapper',
    optimizer=dict(type='AdamW', lr=0.0001, weight_decay=0.05),
    clip_grad=dict(max_norm=10, norm_type=2))
param_scheduler = dict(type='PolyLR', begin=0, end=512, power=0.9)

custom_hooks = [dict(type='EmptyCacheHook', after_iter=True)]
default_hooks = dict(
    checkpoint=dict(
        interval=16,
        max_keep_ckpts=1,
        save_best=['all_ap_50%', 'miou'],
        rule='greater'))

load_from = 'work_dirs/tmp/instance-only-oneformer3d_1xb2_scannet-and-structured3d.pth'

train_cfg = dict(type='EpochBasedTrainLoop', max_epochs=512, val_interval=16)
val_cfg = dict(type='ValLoop')
test_cfg = dict(type='TestLoop')

I don't know how to fix this erro, so I really need some help. Expecting for the solution, thx!

oneformer3d-contributor commented 3 months ago

Please debug smth like for b in YourDataset: print b. Because now you have a problem with dataset. It tries to get i-th element 1000 times, all of the attempts fail (probably bug with path to point clouds or to annotation), so you get this error.

zhaohongxiang-seu commented 3 months ago

@HitmansGO I encountered the same issue, have you resolved it?

ustbwang1jie commented 2 months ago

@HitmansGO have you resolved it?

Lizhinwafu commented 2 months ago

[> @HitmansGO have you resolved it?

](https://github.com/filaPro/oneformer3d/issues/9#issuecomment-2249844455)

clawCa commented 1 month ago

Encountered the same problem. It seems like there was an error in the dataset preprocessing. Did you use the conversion script from mmdetection3d for the S3DIS dataset? In that script, you would need to modify instance categories such as sofa and table to your custom categories.

accoumar12 commented 3 weeks ago

I had the same issue. And solved it by taking care of the good format of the input data.